| svm.select {StabPerf} | R Documentation |
Uses a suppport vector machine (SVM) with a linear kernel (from a customized e1071) as the basis for a greedy feature selection algorithm. The normalized feature weights are used to sort the feautures according to relevance.
svm.select(data, labels, best=NULL, thresh=0.01, start.indices=1:dim(data)[2],...)
data |
matrix or data.frame. Features in columns, samples in rows |
labels |
factor or integer. Labels of the samples of data |
best |
integer. How many features to return, up to total number of feaures. |
thresh |
numeric. Minimum score for a feature to be accepted. |
start |
integer. Reduce search space to these features in data (e.g. from a t-test) |
... |
Other parameters to pass to svm |
Implements a greedy (i.e. non-optimal) feature selection algorithm, which trains a linear SVM and pulls the features based on their normed weights, in order.
If best is given, the best best features are returned. If thresh is given, then all features exceeding this threshold are returned.
Note: This interface depends upon a modified version of the e1071 library (version 1.5-8), which was adapted to incude the features weights and normalized feature weights in the model returned by svm. The respective vectors are stored in the fields fweights and nfweights.
features. list. Selected features. |
fets <- svm.select(t(expr_data), some.factors, thresh=0.01)